Due_Diligence_4DS2 is an intelligent platform that automates and enhances the due diligence process for digital assets and crypto investment funds. Developed as a capstone project by 4DS2 students at ESPRIT University, it leverages advanced NLP, data engineering, and machine learning to deliver a trustworthy, automated due diligence pipeline.
In the volatile and fast-moving crypto landscape, due diligence is critical. This platform enables investors, analysts, and researchers to assess crypto projects with greater accuracy, consistency, and speed.
🔍 From entity extraction and risk profiling to AI-generated Q&A and investor-ready reports — Due_Diligence_4DS2 offers an end-to-end intelligent due diligence solution powered by Retrieval-Augmented Generation (RAG) and Knowledge Graphs.
-
📡 Automated Data Ingestion
Collects structured and unstructured data from APIs, PDFs, websites, GitHub, and regulatory databases using crawlers and pipelines. -
🧠 Custom NLP & Risk Profiling
Uses domain-specific Named Entity Recognition (NER), smart chunking, and multi-category risk tagging (legal, technical, financial, operational) withspaCy, regex, and blacklists. -
🔎 Hybrid Retrieval & Knowledge Graphs
Combines semantic vector search (Qdrant) with keyword-based TF-IDF scoring. Results are enriched using a semantic knowledge graph for multi-hop reasoning. -
🤖 AI-Powered Question Generation
Uses Retrieval-Augmented Generation (RAG) with Qdrant indexing to fetch context from legal, financial, and technical documents. Responses are generated using locally deployed LLaMA 3.2 models, ensuring privacy, auditability, and domain specificity. -
📊 Professional Report Generation
Automatically produces clear, well-structured due diligence reports and PowerPoint slides across categories like Legal, Financial, Technical, Governance, ESG, Risk, and more.
Languages & Frameworks:
Python · JavaScript (React) · FastAPI
ML & NLP:
LLaMA 3.2 1B (local) · Sentence Transformers (MPNet) · Cross-Encoder · spaCy (custom NER)
Data & Pipelines:
Apache Airflow · LangChain · Qdrant · TF-IDF · SQLite · Pandas · NumPy
Document & Web Scraping:
PyMuPDF · PDFMiner · BeautifulSoup · Selenium · GitHub API · Reddit API · CoinGecko · CryptoPanic
Retrieval & Indexing:
Qdrant · Hybrid Search (semantic + keyword) · Smart Chunking · Knowledge Graph
Visualization & Reporting:
python-pptx · Matplotlib · Seaborn
CI/CD & Hosting:
GitHub Actions · GitHub Pages
Evaluation uses RAGAS, covering:
- 🔍 Retrieval Quality – Context precision, recall, and noise sensitivity
- ✍️ Answer Generation – Factuality, completeness, and semantic alignment
- 📊 NLP Metrics – BLEU, ROUGE, BERTScore, Exact Match
Designed for auditability, explainability, and use in high-stakes financial domains.
- 🚀 Deployed using FastAPI
- 🧠 LLaMA 3.2 runs locally (GPU or CPU)
- 🔍 Document vectors indexed via Qdrant only
- 📄 Reports generated with python-pptx
Developed by 4DS2 students at ESPRIT University, under faculty mentorship.